Azure Databricks Compute Options

1
SERVERLESS COMPUTE
2–6 sec startup
Fully managed
Scale to zero
Performance-optimized
1

Serverless Compute

Fully managed, on-demand Databricks compute that starts quickly, automatically scales with workload demand, and removes most cluster management tasks from the user.

2
CLASSIC COMPUTE

STANDARD

Multi-user Python / SQL / Scala

DEDICATED

Single user or group RDD / R / GPU
3–7 min startup
2

Classic Compute

User-provisioned compute that runs in your Azure subscription. Use it when you need custom networking, special libraries, GPUs, R workloads, or more direct control over cluster configuration.

3
SQL
WAREHOUSES

SERVERLESS

2–6 sec Photon + Predictive IO + IWM

PRO

~4 min Photon + Predictive IO

CLASSIC

~4 min Photon only
3

SQL Warehouses

Compute resources optimized for SQL analytics, dashboards, and BI tools. Serverless is fastest and most automated; Pro and Classic provide alternatives when serverless is unavailable or custom networking is required.

4
INSTANCE POOLS
<1 min startup
Pre-warmed VMs
Pay for idle capacity
4

Instance Pools

Reusable groups of pre-warmed virtual machines that reduce cluster startup time. They improve responsiveness, but you pay for idle capacity while VMs wait in the pool.

5
JOB COMPUTE

SERVERLESS JOBS

2–6 sec Auto-terminated

CLASSIC JOBS

3–7 min Spot instances
5

Job Compute

Compute created specifically to run automated workflows and tasks. It is preferred for production jobs because it can start for a run, execute the workload, and terminate afterward to control cost.

Click any row to flip it and see the definition.